Adaptation and sustainability of water management for rice agriculture in temperate regions: The Italian case‐study

We review, analyse, and discuss the recent evolution and the future sustainability of rice paddy fields in Italy—the largest European producer—using outcomes from available literature and new analysis of agricultural statistics from local authorities, land‐use and surface temperature data from remote sensing, hydrological and climate data from observations, and numerical models. We show that Italy can be considered a good representative for rice cultivation in temperate regions that are not freshwater‐limited. However, this situation is changing. We report strong evidence linking the largest European reduction of seasonal surface water that have gradually occurred since 2000 over the rice cultivation area of Northern Italy, to the change in paddy management from traditional continuous flooding to a less greenhouse‐gases‐emitting practice, that is, dry‐seeding with postponed flooding. This change was accompanied by several improvements in agro‐practices and crop varieties. Concurrently, regional climate rapidly shifted towards sunnier weather conditions that partly contributed to higher rice yields and stability, decoupling yields from inter‐annual climate variability, but also reducing water availability. In Northern Italy, a complete shift of rice cultivation towards dry‐seeding is not compatible with seasonal water availability, and a number of drawbacks, with respect to the traditional wet seeding, are also identified from literature review. Therefore, in the context of near‐term climate change, sustainable rice cultivation in the middle latitudes seems achievable (without limiting production and/or increasing volatility) by balancing traditional and dry‐seeding.


| INTRODUCTION
The Po Valley in Northern Italy is characterized by sunny summers typical of the Mediterranean climate and large water availability from spring precipitation and runoff from the surrounding mountain ranges (Zampieri, Giorgi, Lionello, & Nikulin, 2012;Zampieri, Scoccimarro, Gualdi, & Navarra, 2015). These unique conditions, as well as the highly developed irrigation infrastructures (see Figure 1), has promoted Italy as the top European rice supplier with a production of almost 1.6 million tons of paddy in 2016 (FAOSTAT, 2018).
Italian rice is sown between the end of April and the end of May; it flowers in July, and it is harvested in September/October. During the growing season, it needs between 1,500 and 3000 mm of water for irrigation depending on the year . Rice represents the most profitable cultivation in Northern Italy, but it also has the largest impact in terms of fertilizer and pesticides' use (Bechini & Castoldi, 2009). It is mostly grown in the upper Po Valley (see The Po basin hydrological cycle is subject to large inter-annual variability related to the meteorological conditions (Montanari, 2012;Zampieri et al., 2016) that strongly affects crop yields (Ceglar et al., 2018;Zampieri et al., 2019;Zampieri, Ceglar, Dentener, & Toreti, 2017). During sunny seasons that are favourable for rice, less rain feeds rivers and channels, the watertable is lower, and larger withdrawals are needed for rice paddy fields irrigation, especially in Lombardy fields where sandy soils drain faster than in Piedmont.
In the temperate regions of the middle latitudes, dry-seeding with postponed flooding can be implemented to save water, save seeds, reduce labour and machinery cost, and simplify the management.
However, this practice can be used when night-time temperatures are not too low (Hill, Buyer, Bocchi, & Clampett, 1991;Sipaseuth et al., 2007). In Northern Italy, low temperatures during the establishment (Ranghetti et al., 2016;Russo & Callegarin, 1997) and especially during the flowering stages (Russo & Callegarin, 1997) can also be limiting factors for rice yields. Dry-seeding was tested in Italy already in the late 1980s in order to improve the establishment in sandy soils and reduce fermentation of organic matter and infestation by aquatic weeds (Moletti, Giudici, Nipoti, & Villa, 1990). However, heavy rain events were found to hamper field preparation for dry-seeding and increase diseases risk afterwards (Moletti et al., 1990). Over the last 20 years, dry-seeding has been continuously developed and FIGURE 1 Study region in Northern Italy. Yellow areas represent rice paddy fields. Brown lines represent the artificial channels indicating the level of anthropization. The red dot corresponds to the location of the Po River discharge monitoring station of Piacenza. The dashed box indicates the region where the analysis is focused on, that is, the Vercelli and Novara provinces in Piedmont and the Milano and Pavia provinces in Lombardy. It is worth to notice the Cavour Channel (in red), built in the 1860s, linking the Po River in the province of Torino to the Ticino River in the province of Novara [Colour figure can be viewed at wileyonlinelibrary.com] increasingly adopted in Italy (Ente Nazionale Risi, 2014;Ranghetti et al., 2016;Ranghetti, Cardarelli, Boschetti, Busetto, & Fasola, 2018). Compared with the traditional water seeding practice, dryseeding in Northern Italy does not reduce yields , whereas it reduces the irrigation demand of about the 20% (Cesari de Maria et al., 2017) and the global warming potential of 56% with respect to continuously flooded rice paddy fields (as the reduction of methane outbalances the increase of nitrogen oxide emissions, Peyron et al., 2016). This paper synthetizes an extensive literature review, combined with results from field experience and original analyses relevant for rice cultivation sustainability in Northern Italy, considered a good representative for the typical conditions found in the rice growing areas of the middle latitudes.
First, we introduce the study region, and we show the changing distribution of flooded rice field in May through satellite data (Section 3.1).
Then we present a classification of the world's largest rice producers in terms of the hydro-climatological drivers of inter-annual yield variability (Section 3.2), using results from observations and from hydrological model simulation driven by observed data.
We investigate the past and current links between Italian rice yield anomalies and downstream river discharge inter-annual variability (Section 3.2) as well as the regional yield sensitivity to local climate variability (Section 3.3) using observed data in two different periods (1975-1994 and 1996-2016).
We estimate current and future climatic trends over rice paddy fields in Northern Italy by analysing observations and climate model simulations (Section 3.4). We also show preliminary results of customized regional climate model experiment and surface brightness temperature analysis estimating the effects of rice flooded fields on local climate.
We finally conclude synthetizing our results in concert with the available literature (Section 4.1) and discussing the possible issues for future sustainability of rice paddy fields in Northern Italy (Section 4.2).

| Study region and changing distribution of dry and wet seeded rice paddy fields
The rice field spatial distribution map ( Figure 1) has been produced with CORINE 2012-layer 213 (rice paddy fields) data. The spatial distribution of irrigation channels has been provided by regional The analysed Piedmont and Lombardy regions pixels are the ones classified as rice paddy fields in the CORINE land use maps. The ratio of noise (due to cloud contamination), over the analysed areas and for the analysed years, is less than 1%. Details on the exploited LANDSAT images are reported in Table 1.
With reference to the local rice crop calendar, we selected satellite acquisitions at the end of May. In this period, paddy fields are in different conditions according to farmers' management: already flooded in the case of traditional sowing management or still dry in the case of dry sowing and delayed field submersion (locally called 'dry sowing' technique).
NDFI is derived from satellite spectral bands sensed along the red and short wave infra-red regions and is considered one of the most sensitive approach to identify standing water. Therefore, it is often exploited in remote sensing as a proxy of rice paddy fields flooding conditions (Boschetti et al. 2014). The detection of rice flood is performed by applying a threshold value of 0.32 to the index. Such a value was identified by Ranghetti et al. (2016) as optimal in discriminating flooded and non-flooded pixels and for the creation of high resolution reference flooding maps for the case-study of Northern Italy

| Global analysis of hydro-climatic drivers of yield variability
The analysis of rice yield and climate variability is conducted at country level by using FAOSTAT data (FAOSTAT, 2018; www.fao.org/ faostat/en/) and non-parametric estimators of heat stress and soil moisture anomalies over rice paddy fields, following a recently proposed methodology applied in several other studies Zampieri et al., 2019;Zampieri, Ceglar, et al., 2017) Heat stress and drought/water excess are estimated, respectively, by the Heat Magnitude Day (HMD, Zampieri, Ceglar, et al., 2017) and the Standardized Precipitation Evaporation Index (SPEI, Vicente-Serrano et al., 2013). The SPEI is a proxy for soil moisture inferred by the local surface water balance (e.g., precipitation minus potential evapotranspiration).
These two indicators are computed on global gridded datasets derived from climate observations and available from 1980 to 2010.
They are evaluated for each grid point containing rice paddy fields and using global crop distribution and calendar information from the MIRCA2000 dataset (Portmann, Siebert, & Döll, 2010) for the identification of the (crop-dependent) higher climate sensitivity period.
MIRCA2000 dataset includes multiple cropping seasons, which can be quite important for rice especially in the tropics . The HMD is computed in the last month before harvesting.
The SPEI is evaluated in the last 1, 2, or 3 months before harvesting, automatically choosing the configuration that maximises the skill (as in Ceglar et al., 2018).
We aggregate the HMD and SPEI time-series at country level, and we isolate the inter-annual anomalies of yield, HMD, and SPEI from the baseline trend through a non-linear trend estimation with locally weighted scatterplot smoothing (Cleveland & Devlin, 1988).
The bilinear combination of the standardised anomalies of HMD and SPEI, that is, the Combined Stress Index (CSI), is an estimator of the yield anomalies. The CSI is built with a bilinear ridge regression at country level by using yield data from FAOSTAT (www.fao.org/ faostat).
An alternative version of the CSI replaces the SPEI with an indicator of non-local water transport, that is, the Standardised River Discharge Index (SRDI), which is a proxy for surface water availability computed on a global gridded hydrological simulation driven by observations (see Zampieri et al., 2018). At the global level, the CSI with the SPEI explains 6.7% of rice global production variance, whereas the CSI with the SRDI explains the 32% of rice global production variance.
The comparison at country level, in terms of explained yield variability, of the CSI-SPEI and the CSI-SRDI identifies where rice is more sensitive to the local precipitation and evapotranspiration balance versus the non-local water balance determined in the basin upstream. The sign of the regression coefficients corresponding to the SRDI allows distinguishing the countries more prone to drought with respect to the countries more affected by water excess and related factors such as cloudiness and precipitation.

| Yield-river discharge relationships in Northern Italy
Discharge data of the Po River in Piacenza have been downloaded from the Arpa Emilia Romagna website (www.smr.arpa.emr.it/ dext3r/, accessed on 13/11/2018).
Po River discharge in Piacenza is compared with FAOSTAT rice yield data. This spatial scale difference does not compromise the analysis because most of the national rice yield production is carried on in Piedmont and Lombardy, upstream of Piacenza. Two well-distinct periods, with different correlation between yield and river discharge anomalies, are identified.
These two periods are further investigated with the observed meteorological data retrieved from the EC-JRC MarsMet Archive (Biavetti et al., 2014) and available at http://agri4cast.jrc.ec.europa.eu/. Climate indicators are computed with the meteorological data over rice paddy fields, aggregated at the province level, and related to yield data for Piedmont and Lombardy provided by Ente Nazionale Risi. Because FAOSTAT data at country level are derived from Ente Nazionale Risi data, the two datasets are consistent.
Non-linear trends are removed from the yield, river discharge, and climate indicators time-series before computing the linear correlations. Statistical significance is computed according to a two-sided t test (p < .1).

| Observed and projected climate change
Climate change projections are computed combining an ensemble of four high-resolution (i.e.,~11 km horizontal resolution) simulations contributing to the EURO-CORDEX Initiative (http://www.eurocordex.net/). This ensemble includes the CNRM-CCLM4 regional climate model driven by boundary conditions from the CNRM-

| Climate effects of dry-seeding estimated by regional climate modelling experiment
We design a simplified regional climate model experiment to provide a first estimate of the surface temperature response to paddy field flooding by using the WRF-ARW model version 3.6.1 (Shamarock et al., 2008). This model is implemented on a mesh composed by 100 × 100 grids of 10 × 10 km horizontal resolution, centred at 45°N and 10°E, and having 36 vertical levels with variable resolution, increasing close to the surface. The model parameterizations include the Noah land-surface model (Niu et al., 2011).
The model is initialised and driven by ERA-Interim reanalysis data (Dee et al., 2011) for 2003. The spin-up run starts the 1st of January.
In April and May, two experiments are performed using different land cover categories over the area where the biggest reduction of flooding has been observed by using LANDSAT data ( Figure 2). The 'inland water' category is used for the control run, corresponding to flooded areas in the period before 2000. Scenario model run is performed by replacing the above mentioned land use category with mixed cropland, as a proxy for conditions imposed by dry sowing in the rice paddy fields.

| Climate effects of dry-seeding estimated by satellite data
To obtain an estimation of the temperature change induced by the change in management practice from traditional to dry-seeding, we analyse satellite time-series of land surface temperature (LST) over designated areas in the study zone.
The first step consists in identifying plausible fields in which either dry-seeding or conventional seeding has occurred. To do this, we use a series of Sentinel-2 multispectral imagery with a spatial resolution of 10 m to detect the plausible dates of the first observable flooding  For each 1-km grid cell of this MODIS dataset, we calculate the fraction covered by either dry-seeding and traditional seeding, as identified from the Sentinel-2 images in the previous step. Because fields are usually smaller than 1 km 2 and not all fields can be identified, these fractions may underestimate the real proportion of the signal coming from a given rice management type. In absence of any systematic bias, we can assume that dry and traditional seeding are estimated with equivalent degrees of confidence. Time-series in which the proportion of dry sowing is larger than that of traditional sowing by more than 10% are considered to be dominated by dry-seeding, whereas the reverse is done to isolate population representing traditional sowing.  The reduction of surface water extent related to dry-seeding in Northern Italy is also evident from the analysis of individual satellite images in order to specifically diagnose the flooded rice paddy fields (Boschetti et al., 2014;Ranghetti et al., 2016Ranghetti et al., , 2018 Ceglar, et al., 2017), for the period 1980-2010. In one case, the CSI includes an indicator of river discharge anomalies (Standardized River Discharge Index, SRDI, Zampieri et al., 2018). In the other one, the CSI includes an indicator for local water balance, that is, the Standardized Precipitation Evapotranspiration Index (SPEI, Vicente-Serrano et al. 2010). The x-axis shows the regression coefficient associated to the SRDI. Both the CSI-based models include the same heat stress indicator. Thus, differences can be attributed to the different sensitivity of rice yields to the upstream runoff with respect to the local water balance over rice-cultivated regions. Labels correspond to ISO 3166 country codes. Red lines delimit the quadrants of the scatter plot [Colour figure can be viewed at wileyonlinelibrary.com] substantially improve by considering the non-local water balance indicator.

| Study area and change of surface waters due to rice dry-seeding
In China, India, Nepal, Bangladesh, and Thailand, yield variability is better captured by the SRDI, and the sign of the regression coefficient is positive (top right quadrant in Figure 3). This is the expected behaviour for irrigated rice in paddy fields where freshwater availability represents a limiting factor.
A third set of countries (Uruguay, Italy, Argentina, North Korea, and the United States) is better correlated to the SRDI than to the SPEI, but the sign of the regression coefficients is negative (top left quadrant in Figure 3). In these countries, rice paddy fields are not freshwater limited. Thus, high river discharge could be considered as a proxy of cold and wet seasons that are unfavourable for the vegetative phase. More detailed analysis using the observed river discharge data for Italy ( Figure 4) from 1971 to 2016 also points to a statistically significant anti-correlation with rice yield anomalies (Figure 4b; r = −.64, p value < .01). This correlation is stronger before 1995, but it depends on the integration period of river discharge (Table S1). The

correlation starts being significant when May is included and increases
when the period is extended until October. In the recent period, yield anomalies are correlated with the river discharge in October only.
The maximum yield loss in the available data occurred in 1977 with an anomalies of 1.5 tons/ha (28% of the baseline value). In that year, the maximum river discharge anomaly was observed with an integrated value of about 18 billion m 3 from June to September During the period 1971-2016, yield has steadily increased from 5 tons per hectare to almost 7 tons per hectare (Figure 4a). At the same time, river discharge data exhibit an overall significant negative trend during the rice-growing season (Table S2). Nevertheless, yield and river discharge trends are not related with each other, as yield increase is mainly due to improved management, varieties, and CO 2 fertilization effect while water availability was not a limiting factor (see Section 4).

| Rice yields and climate variability in Northern Italy
For Italy, we have found a sharp separation between two periods, suggesting that the sensitivity of yield anomalies to inter-annual climate variability suddenly disappeared in the mid-1990s (Figure 4). In order to explain the decoupling of rice yield and river discharge, we analyse the links between regional rice yields and meteorological indicators in the two 21-year periods (1975-1995 and 1996-2016) computed with the meteorological dataset MarsMet (see Section 2). Figure  Increased sensitivity to minimum temperature is found during May in Piedmont after 1995. This feature could be related to changes in varieties or agro-management rather than to changes in climate (see Section 4.1). However, available data do not allow to fully explain it.
In Lombardy, after 1995, yields become positively correlated to temperature until September (consistently with the hypothesis that the crop is favoured by sunnier climate) and to precipitation in June (consistently with the additional water required by delayed flooding).
At the end of the growing season, in October, yields are negatively affected by warm temperature anomalies in Piedmont and especially in Lombardy, probably because of the shortening of the grain filling period. These relationships do not change between the two periods.
In the next section, we perform a deeper analysis of the climate change signal over rice field, shading light on some of the results obtained on the correlations between climate and yield inter-annual variability.

| Climate change over rice paddy fields in Northern Italy
Rice paddy fields in north-western Italy are characterised by a bimodal precipitation annual cycle with two peaks in spring and autumn of about 100 mm and large inter-annual variability (Figure 6a). Summers tend to be drier, as a reminiscence of the Mediterranean climate in this transition region (Zampieri et al., 2012;Zampieri et al., 2015). The seasonal cycle of maximum and minimum temperatures reaches its unique peak around July with 29°C and 18°C, respectively.  periods 1975-1995 and 1996-2016. In the last 21 years, a significant reduction of more than 20 mm of rain is found in May, July, September, and October. This drying trend very likely explains the reduction of sensitivity to high precipitation and low temperatures found in Piedmont and into a lesser extent in Lombardy.
Temperatures warmed up significantly during all the growing season. In spring, we diagnose almost 2°C warming during the day.
Minimum temperatures passed from 6.7 ± 0.9°C to 7.9 ± 1.0°C in April from 10.7 ± 1.3°C to 12.2 ± 1.0°C in May, surpassing the minimum critical threshold for the suitability dry-seeding in cold climates (Sipaseuth et al., 2007).
The projected changes of precipitation are not significant.
Preliminary analysis of numerical simulations conducted with the regional climate model shows that the land-use change corresponding to dry-seeding contributed in the opposite way, especially during the night (see Figure 7). In particular, numerical experiments suggest that total conversion to dry-seeding would have resulted in a change of surface energy balance producing an average cooling of (−0.81 ± 0.31°C). The cooling is larger and more stable during the night (−1.1 ± 0.49°C). During the day, the cooling is reduced and more variable FIGURE 5 Estimated correlation between selected agro-climatic indicators (JRC-EC data) and rice yields in Piedmont (P) and Lombardy (L) before 1995 (B: 1975-1994) and after 1995 (A: 1996-2016). Top labels represent, from left to right, the selected climate indicators: maximum (tmax), minimum (tmin) and average (tavg) temperature; accumulated precipitation (rr); number of days with precipitation above 10 mm (ndays. (−0.74 ± 0.71°C). Thus, the diurnal temperature range at 2 m is increased by dry-seeding, consistently with the expectations. The cooling signal due to dry-seeding was completely outbalanced by the regional rapid warming recorded in Northern Italy.
The effect of dry-seeding on land surface temperatures (

| Concurrent changes accompanying dry-seeding in Italy
The combined changes affecting the rice agricultural system and the environment in the last 40 years are shown in Figure 9, grouped by the main aspects of the agricultural, environmental and climate system:

| Rice yield
Increasing yields and decreasing sensitivity to climate characterize the general evolution of rice cultivation in Northern Italy (Figure 4).
Unfavourable weather conditions (i.e., cold and rainy days) were negatively affecting yields before 1995 ( Figure 5). These conditions are common to the other main rice producers of the middle latitudes ( Figure 3, upper left panel). Amongst other factors, rice yields increase because of the CO 2 fertilization effect of about a 0.088%/ppm (Ainsworth, 2008) to 0.095%/ppm (Kimball, 2016). Atmospheric CO 2 concentration passed from 330 ppm in 1970 to the current value of about 410 ppm (407 ppm in 2016), so it could be responsible for up to 6.8-7.3% increase of yield. Therefore, one fourth of the observed rice yield increase in Italy could be due to the increase of CO 2 . The remaining portion is presumably due mostly to improved varieties and management.

| Climate
The region was characterised by a rapid warming and drying in spring and summer because of natural climate variability (i.e., positive shift of the Atlantic Multidecadal Oscillation occurred in the mid-1990s (Zampieri, Toreti, Schindler, Scoccimarro, & Gualdi, 2017) superimposed to the general global warming. The change in summer was very likely favourable for rice yields. In spring, preliminary results show that dry-seeding land-use change is associated to a significant amplification of the diurnal temperature range and to a general cooling of the region, albeit smaller than the warming due to natural and anthropogenic climate change (Figures 6 and 7). If these results are confirmed, it would mean that dry-seeding partly compensated the warming produced by climate change and the change of circulations patterns. The negative temperature feedback associated to dryseeding land-use change is represented by the solid upward arrow in Figure 9. The AMO shift in spring is also associated to larger rain versus snow ratio (Zampieri, Scoccimarro, & Gualdi, 2013) in the surrounding mountains and to the shift towards earlier river discharge (Zampieri et al., 2015).

| Agromanagement
The main change is constituted by the adoption of dry-seeding with no negative effects on yields and yields variability . This transition was accompanied by development in varieties and agromanagement practices (seeds amount, seeding patterns, fertilizations, etc.; Moletti et al., 1990). Land levelling with laser-equipped machine has been also implemented in the field preparation phase.
Weed control has been evolved introducing new products (as only water-soluble products were available in the 1980s). The new varieties, however, require more fertilizer and are less competing with weeds.
Since 2006, new genotypes selected in Louisiana that are less resistant to shock by low temperature in microsporogenesis started spreading and now cover a surface of 70,000 ha. Dry-seeding cropping system in Northern Italy decreases emission of methane but increase those of nitrogen oxide, with a combined effect of reducing the global warming potential of 56% . This effect is represented by the dashed arrow in Figure 9.

| Water management
River discharged during the rice growing season decreased in the last 40 years as a result of drying climate (Zampieri et al., 2013;. Dry-seeding saves about 20% of water at the beginning of the growing season when the water abundance in the river is still high. However, it requires larger irrigation in June to raise the water table and flood paddy field. This coincides with a period of lower water availability (Zampieri et al., 2015) and higher water competition with other crops such as maize (Zampieri et al., 2019). This factor constitutes the main limitation for the sustainability of dry-seeding adoption in Italy, which has already reached its maximum potential in this respect.

| Environment
Increase in environmental pollution from herbicides, needed in the dry-seeding practice (Rao, Johnson, Sivaprasad, Ladha, & Mortimer, 2007), is likely, but not quantified for Italy yet. On the other hand, high soil solution nitrate concentration is already identified as the greatest environmental constraint of dry-seeding cropping system , jeopardizing river and groundwater quality. However, nitrate concentration in the groundwater is almost everywhere contained below 25 L −1 in Piedmont and Lombardy (EU Knowledge Hub Water and Agriculture https://water.jrc.ec.europa.eu/). An emerging constraint is the arsenic (As) concentration in the soils, as rice is an important pathway for inorganic As dietary intake and the rate of As absorption increases with dry-seeding (Tenni et al., 2017).
However, monitored concentrations are still below the acceptable limits indicated by the EU, and agronomic strategies are being investigated to control and ameliorate food safety (Tenni et al., 2017).
Considering ecosystems, dry-seeding could bring some serious drawbacks as traditionally flooded rice paddy fields provide habitat for spring migrants and locally breeding birds (Imperio, Ranghetti, & Hardenberg, 2017). Noteworthy, the regional administrations (Regione Lombardia and Regione Piemonte) implemented effective alternatives for preserving wetlands ecosystems though 'nature-based' waste water and flood protection plants, albeit at a small spatial scale (Liquete, Udias, Conte, Grizzetti, & Masi, 2016).

| Future perspective
As for the future, climate projections results computed on an ensemble of high-resolution regional climate models suggest a slower warming compared with what has been already observed in the last decades. An emerging climate risk appears in the higher emission scenario after 2050 ( Figure 6) as maximum daily temperature will exceed the 33°C threshold (Luo, 2011), offsetting the positive effect of higher atmospheric CO 2 concentrations (Chaturvedi, Bahuguna, Shah, Pal, & Jagadish, 2017). Albeit mean precipitation changes are not significant ( Figure 6), there is risk of increasing extreme events (Scoccimarro, Gualdi, Bellucci, Zampieri, & Navarra, 2016;Toreti & Naveau, 2015).
However, the main adverse effects of climate changes could be handled by implementing adaptation strategies such as shifting varieties and earlier sowing (Bocchiola, Nana, & Soncini, 2013). Future climate projections suggest no additional constraints on freshwater availability for rice irrigation (Elliott et al., 2014). However, water demand is high in the region (Miglietta, Morrone, & De Leo, 2018), and water resources are subject to multiple pressures, such as nitrogen pollution from agriculture (mainly caused by livestock farm), alteration of the natural river flow regime for water abstractions, contributing to the degradation of the ecological status and biodiversity of rivers (Grizzetti et al., 2017) and reducing the value of the related ecosystem services (Grizzetti et al., 2019).

| Final remarks
Rice cultivation in Italy was subject to profound agro-management changes in the last 40 years (Figure 9), the most prominent being the introduction of dry-seeding practice.
The corresponding 1-month delay of paddy field flooding most probably produced the largest anthropic seasonal surface water reduction detectable in the whole Europe and corresponded to a significant decrease of greenhouse gases emissions .
Local experiments have shown how dry-seeding can save water in the early stages of crop growth (Cesari de Maria et al., 2017) and that dry seeded rice yields are compatible with those ones obtained with the traditional continuous flooding practice . This agro-management change was made possible by several adaptation measures, but it was likely also favoured by rapid climatic change towards sunnier weather conditions that occurred in the mid-1990s . Our analysis also highlighted a sharp decrease of climate influence on yields' variability during the 1990s.
However, rice dry-seeding could endanger sustainable water management in Northern Italy. In fact, the entire mountain ranges surrounding the region experienced rapid warming and drying trend (Zampieri et al., 2013) associated with earlier runoff generation (Zampieri et al., 2015) and decrease of water availability during the rice growing season (see Figure 4, Table S2). In addition, dry-seeding saves water at the beginning of the growing season (Cesari de Maria et al., 2017) when availability is high, whereas it requires more water later in the season (June/July) when the availability is lower (Zampieri et al., 2015) and the competition within the sector and with other sectors is higher. In this respect, dry-seeding has already reached its maximum potential (Associazione Irrigazione Est Sesia and Associazione Irrigazione Ovest Sesia, "Note for the Ministry of Agriculture, Piedmont Region and Lombardy Region," 10th January 2018).
Furthermore, dry-seeding would increase nitrate water contamination  and disrupt wetland ecosystems (Imperio et al., 2017). For all these reasons, a sustainable balance between dry seeded field and conventionally flooded fields has to be attained.
Our analysis also highlights several issues for sustainability that are not directly related to dry-seeding ( Figure 9). However, public monitoring tools are now provided by the European Union to raise awareness on environmental issues related to water and agriculture (https:// water.jrc.ec.europa.eu/).
We have shown preliminary results diagnosing a significant local climatic effect due to the flooded rice paddy fields that are increasing the diurnal temperature range and cooling down the entire region ( ORCID Matteo Zampieri https://orcid.org/0000-0002-7558-1108